NewRecommendations on your thank you and order status pages, and moreThank you page and moreSee what's new →
3D shopping basket, paper bag and chart icons on a blue background
Educational

What is a product recommendation engine, and should you pick pairs by hand?

June 15, 2026 · 4 min read · Updated October 26, 2026

A product recommendation engine is the software that decides which products to show next to the one a shopper is viewing or buying. It can follow rules you write, count which products are bought together, compare product details or personalize per shopper. A small store has little order history, so it does best with automatic picks from product details, plus hand-picked pairs on its ten best sellers.

Four ways an engine picks products

  • Rules: you decide ("show the case with every phone") or set conditions by tag or collection. Fully predictable, but every rule is something you write and keep current.
  • Bought together: the engine counts which products appear in the same orders. It needs a lot of orders to be reliable, and a new product has no history.
  • Similarity: the engine compares product titles, descriptions, tags and types to find items that are alike or go together. It works from day one with no orders, but it can miss relationships your products don't describe.
  • Personalization: the engine shows different products to different shoppers based on what each one browsed and bought. It needs steady traffic to learn, and a way to test whether it beats simpler methods.

Most apps mix these. When an app says "AI", ask which of the four it means and what data it needs. Shopify's own related products work the same way: purchase history where it exists, product descriptions on English storefronts, related collections otherwise.

What a small store needs

A store with a few hundred orders a month and 10 to 500 products doesn't have enough data for personalization, or for reliable bought-together counts across the whole catalog. With few orders, most pairs have appeared together only a few times, so they're a weak signal on their own until orders build up. What a small store needs is sensible recommendations on day one (similarity), real bought-together pairs where the orders exist, and filters that never show sold-out items or products already in the cart.

Picking pairs by hand or automatically

Picking by hand means choosing which products show with each product, for example up to 10 complementary and 10 related products per product in Shopify's free Search & Discovery app. You know your catalog, so your pairs are usually the most sensible ones. The cost is time: every new product, sold-out item or seasonal change means another round of edits. Automatic recommendations give every product a row, new ones included, with no list to maintain, at the price of some odd picks when product details are thin.

Example, with assumed numbers: at three minutes per product, picking pairs for a 30-product store takes an hour and a half, and some of that again whenever products change. A 300-product store would need fifteen hours for the first pass. The smaller and steadier your catalog, the more hand-picking makes sense.

Most stores do best with both. Let recommendations be automatic across the catalog, then check your ten best sellers by hand and fix any pick that's wrong. Those products get most of the views, so a few minutes there covers most of what shoppers see. Look again when you add a product line or the season changes.

Give the engine better product details

Automatic methods read what you give them. Clear titles, a description that says what the product is for and useful tags ("cast iron", "gift", "refill") make the picks better, for Shopify's related products and for apps alike. The cross-sell ideas by store type list the tag that matters for each kind of store, such as the model for electronics or the species for pet supplies.

Questions to ask before you choose one

  • Does it work with no order history, and what does it show until orders build up?
  • Does it use only your store's data, or data from other stores?
  • Does it leave out sold-out products and items already in the cart?
  • Can you change the picks for a product when one is wrong?
  • What does it count as a sale from a recommendation? The guide to tracking upsell revenue explains why the answer changes the numbers.

How Dropr picks products

Dropr uses only your store's own data. For "You might also like" it compares your products' titles, descriptions and tags to find what goes with the one on screen, and a shared tag ranks a product higher. "Bought together" uses products that appear in the same paid orders since you installed Dropr, and "Popular right now" uses best sellers from those orders; both fill in from your catalog until there are enough orders. "Recently viewed" keeps each shopper's history in their own browser. Sold-out products and items already in the cart are left out. It doesn't personalize per shopper or pool data across stores. On Dropr's Products page you can reorder, remove or add picks for any product, and your own picks are marked "Added by you".

Frequently asked questions

Is "frequently bought together" a recommendation engine?

It's one method an engine can use: counting co-purchases. The guide to frequently bought together explains its limits on a small store.

Will a recommendation engine slow my store?

The engine runs on the app's servers; your store loads the app's script and the product images. The guide to upsell apps and store speed shows how to measure the effect.

Related reading

Try Dropr on your store

Free up to 1,000 views a month. $19 every 30 days for unlimited views.

Install free

No theme file edits · Pricing · What's new